Entity Extraction
Pulling the specific things out of a sentence — the order number, the date, the branch, the amount — so the assistant can act on them rather than merely understand them.
Pulling the specific things out of a sentence — the order number, the date, the branch, the amount — so the assistant can act on them rather than merely understand them.
What it does
Entity extraction identifies the concrete items inside a request. "Has order 4021 shipped to Jeddah yet" contains an order reference and a city, and retrieving anything useful depends on isolating them correctly.
It is the difference between understanding a question and being able to answer it. Intent tells the assistant what kind of question this is; entities tell it which record to look at.
Where it shows up in a conversation
Record lookup
Extracting the reference that identifies which order, case or account is meant.
Dates and periods
Understanding "last Tuesday", "this month" and their Arabic equivalents, including Hijri references.
Amounts
Identifying figures and currencies, including where numeral systems are mixed.
Locations
Branch, city and district names, including the informal variants people actually use.
Product references
Matching what a customer calls something to what your catalogue calls it.
Form filling
Gathering the fields a process needs, conversationally rather than as a form.
Why it is harder in two languages
Arabic entity extraction faces problems English does not. Numerals appear in two systems and documents and messages mix them, so an amount or a reference may be written either way, sometimes within one sentence.
Dates are harder still. Both Gregorian and Hijri calendars are in daily use, relative expressions differ structurally between the languages, and the week begins on a different day — which changes what "next week" means. An extractor that assumes one calendar will produce confidently wrong dates.
Names and places are the third difficulty. Transliteration between Arabic and Latin script is not standardised, so a single district or surname may appear spelt several ways in English. Matching across those variants is a practical requirement rather than a refinement.
What to look for when you evaluate it
| Numeral handling | Both numeral systems, including mixed within one message. |
|---|---|
| Calendar handling | Gregorian and Hijri, and relative expressions in both languages. |
| Transliteration variants | Whether differently spelt versions of the same name or place match. |
| Reference formats | Whether your own reference formats are recognised, including with typos. |
| Partial extraction | What happens when only some required entities are present — asking for the missing one rather than failing. |
| Confidence | Whether an uncertain extraction is confirmed rather than acted on. |
Where it should not be trusted alone
An extracted reference should be confirmed where acting on the wrong one has consequence. A mistyped or misheard order number produces a confident answer about somebody else's order, which is both a service failure and a disclosure problem.
Dates deserve particular care. A misinterpreted relative date — the wrong week, the wrong calendar — produces an answer that is precisely wrong rather than obviously wrong, and neither party is likely to notice.
Extraction should also not be used to fill gaps by inference. Where a required detail is genuinely absent, asking for it is correct; guessing it from context is how an assistant confidently retrieves the wrong record.
How this fits the whole
No single capability answers a question on its own. A conversation passes through several in sequence — understanding what was asked, finding the relevant material, composing a reply, checking it before it is sent — and the quality of the result is set by the weakest link rather than the strongest.
That has a practical consequence for evaluation. Vendors tend to demonstrate their strongest component, and a system with excellent retrieval and weak language handling will demo beautifully and disappoint in production, because the demo was conducted in the language and phrasing the system handles best.
It also explains why grounding matters more than raw capability. A more capable model that answers from its own training rather than from your material produces confident, fluent, unsupported answers — which is a worse outcome than a less capable one that retrieves correctly and declines when it cannot. Capability without grounding increases the cost of being wrong rather than reducing it.
The pieces are described separately here because that is how they are evaluated and configured. They are not experienced separately: a customer asking a question encounters one system, and judges it as one.
Where this sits in a conversation
- Something arrivesA typed message, a spoken sentence, an uploaded document. The first job is establishing what was actually received — which language, which script, whether it is legible, whether it is one question or several.
- It is interpretedWorking out what is being asked rather than what words were used. The same question arrives phrased twenty different ways, in two languages, often mixed, frequently misspelt, sometimes as a fragment.
- Material is retrievedThe relevant part of your own approved content is found. This is where answer quality is largely decided: a correct answer assembled from the wrong source is still wrong.
- A reply is composedIn the language of the question, in the configured tone, from the retrieved material and nothing else. Where the material is silent, the reply says so rather than filling the gap.
- It is checkedBefore sending, the reply is examined for claims the retrieved material does not support and for personal data that should not be repeated back. This runs on every reply.
- It is deliveredAs streamed text, as speech, or as a structured response — and the conversation is recorded so that what happened can be reviewed later.
Why this is built rather than bought in
Every capability described here could, in principle, be obtained by calling somebody else's service. That is how most assistants in this market are assembled, and it is why so many of them cannot answer the question that decides regulated procurements: where does the data go.
When a component is a call to an external provider, every question that passes through it leaves the organisation. That may be perfectly acceptable for a retailer answering questions about opening hours. It is usually not acceptable for a bank explaining a transaction, a hospital confirming an appointment, or a public body handling a citizen case — and the constraint is architectural, so it cannot be configured away later.
The second consequence is control over behaviour. A capability obtained as a service behaves the way the provider decides, changes when the provider changes it, and cannot be tuned for the specific way your customers speak or write. A capability that runs inside your own estate can be adjusted for dialect, for your terminology, and for the particular failure modes your material produces.
The third is continuity. External services are deprecated, repriced and withdrawn. A deployment whose core capability depends on somebody else's roadmap inherits that roadmap, including the parts nobody announced in advance.
Measuring this component honestly
| Measured in which language? | Nearly every published benchmark in this field was developed against English. An overall figure that averages across languages hides exactly the gap that matters in this market. |
|---|---|
| Measured on what input? | Clean or realistic. Read speech or spontaneous speech. Typed text or phone-typed text with autocorrect. The difference between these is usually larger than the difference between vendors. |
| Measured by whom? | Self-reported figures with an undescribed method are marketing. Ask for the test set and the procedure. |
| What counts as correct? | A surprising number of published figures count a partially correct answer as correct. Ask what the scoring rule was. |
| How does it fail? | A component that degrades gracefully is worth more than one with a higher headline figure that fails silently and confidently. |
| Does it report uncertainty? | A component that surfaces low confidence lets the system confirm rather than guess. One that hides it produces confident errors downstream. |
What to ask about this specifically
Ask for the Arabic figure
Not the overall figure. Nearly every published benchmark in this field was developed against English, and the gap between the two languages is where deployments disappoint.
Ask how it was measured
Against what test set, by whom, on what kind of input. A number without a described method is marketing rather than measurement.
Ask what it does when uncertain
Whether uncertainty is surfaced or hidden. A component that hides low confidence produces confident errors downstream.
Ask about real conditions
Noisy audio, poor scans, dialect, informal phrasing, mixed languages. Laboratory conditions are not predictive of contact-centre reality.
Ask where processing happens
For any component touching personal data, this is a procurement question rather than a technical curiosity in regulated sectors.
Ask what is retained
What data this component keeps, for how long, and whether it is used to improve anything.
Common questions
It has to. Both calendars are in daily use here, and an extractor that assumes one produces confidently wrong dates that nobody notices.
Both numeral systems are handled, including messages that mix them, which is common.
Your formats are configured during deployment, including tolerance for the typos people actually make.
Transliteration is not standardised, so matching across variants is a practical requirement. A district may legitimately appear several ways in English.
It asks for that one thing rather than presenting a form or failing. Guessing from context is how the wrong record gets retrieved.
Yes — the same capability applied to uploaded files, which is how a form can be checked for completeness.